Target Fitting Method for Spherical Point Clouds Based on Projection Filtering and K-Means Clustered Voxelization.

Area of Science:

• Metrology and Measurement Science
• Computer Vision and Image Processing
• Geometric Modeling

Background:

• Industrial computed tomography (CT) is crucial for non-contact, high-precision measurements.
• Point cloud data from CT scans often contain artifacts, leading to inaccurate geometric parameter fitting.
• Existing methods like least squares and RANSAC have limitations in accuracy and robustness due to noise and artifacts.

Purpose of the Study:

• To develop a robust and accurate spherical point cloud fitting algorithm for industrial CT measurements.
• To address the challenges posed by artifacts and noise in CT-derived point clouds.
• To improve the reliability of geometric parameter extraction in metrology applications.

Main Methods:

• Proposed a novel algorithm, Projection and K-Means RANSAC (PK-RANSAC), integrating projection filtering and K-Means clustering.
• Utilized RANSAC for initial parameter estimation, followed by 2D projection for center coordinate correction.
• Employed K-Means clustering on sampled points and weighted the largest cluster for final parameter determination.

Main Results:

• PK-RANSAC achieved a sphere center fitting deviation of 1.91 μm on a standard ball-plate.
• This represents a significant improvement over the traditional RANSAC method, which yielded a deviation of 25.41 μm.
• The experimental results validate the superior accuracy and robustness of the proposed PK-RANSAC algorithm.

Conclusions:

• The PK-RANSAC algorithm offers a substantial advancement in fitting geometric parameters from noisy point clouds in industrial CT.
• The integration of projection filtering and K-Means clustering effectively mitigates the impact of artifacts.
• PK-RANSAC demonstrates high accuracy and robustness, making it suitable for precise metrology applications.